Mitigating Machine Learning Bias Between High Income and Low-Middle Income Countries for Enhanced Model Fairness and Generalizability (preprint)

Collaborative efforts in artificial intelligence (AI) are increasingly common between high-income countries (HICs) and low- to middle-income countries (LMICs). Given the resource limitations often encountered by LMICs, collaboration becomes crucial for pooling resources, expertise, and knowledge. Despite the apparent advantages, ensuring the fairness and equity of these collaborative models is essential, especially considering the distinct differences between LMIC and HIC hospitals. In this study, we show that collaborative AI approaches can lead to divergent performance outcomes across HIC and LMIC settings, particularly in the presence of data imbalances. Through a real-world COVID-19 screening case study, we demonstrate that implementing algorithmic-level bias mitigation methods significantly improves outcome fairness between HIC and LMIC sites while maintaining high diagnostic sensitivity. We compare our results against previous benchmarks, utilizing datasets from four independent United Kingdom Hospitals and one Vietnamese hospital, representing HIC and LMIC settings, respectively.

SARS-CoV-2 sequencing with cloud-based analysis illustrates expedient co-ordinated surveillance of viral genomic epidemiology across six continents (preprint)

Viral sequencing has been critical in the COVID-19 pandemic response, but sequencing and bioinformatics capacity remain inconsistent. To examine the utility of a cloud-based sequencing analysis platform for SARS-CoV-2 sequencing, we conducted a cross-sectional study incorporating seven countries in July 2022. Sites submitted sequential SARS-CoV-2 sequences over two weeks to the Global Pathogen Analysis Service (GPAS). The GPAS bioinformatics cloud platform performs sequence assembly plus lineage and related sample identification. Users can share information with collaborators while retaining data ownership. Seven sites contributed sequencing reads from 5,346 clinical samples, of which 4,799/5,346 (89.8%) had a lineage identified. Omicron lineages dominated, with the vast majority being BA.5, BA.4 and BA.2, commensurate with contemporary genomic epidemiological observations. Phylogenetic analysis demonstrated low within-lineage diversity, and highly similar sequences present in globally disparate sites. A cloud-based analysis platform like GPAS addresses bioinformatics bottlenecks and facilitates collaboration in pathogen surveillance, enhancing epidemic and pandemic preparedness.

Kinetics of neutralising antibodies against Omicron variant in Vietnamese healthcare workers after primary immunisation with ChAdOx1-S and booster with BNT162b2 (preprint)

We studied the development and persistence of neutralising antibodies against SARS-CoV-2 ancestral strain, and Delta and Omicron (BA.1 and BA.2) variants in Vietnamese healthcare workers (HCWs) up to 15 weeks after booster vaccination. We included 47 HCWs with different pre-existing immune statuses (group 1 (G1): n=21, and group 2 (G2): n=26 without and with prior breakthrough Delta variant infection, respectively). The study participants had completed primary immunisation with ChAdOx1-S and booster vaccination with BNT162b2. Neutralising antibodies were measured using a surrogate virus neutralisation assay. Of the 21 study participants in G1, neutralising antibodies against ancestral strain, Delta variant, BA.1 and BA.2 were (almost) abolished at month 8 after the second dose, but all had detectable neutralising antibodies to the study viruses at week two post booster dose. Of the 26 study participants in G2, neutralising antibody levels to BA.1 and BA.2 were significantly higher than those to the corresponding viruses measured at week 2 post breakthrough infection and before the booster dose. At week 15 post booster vaccination, neutralising antibodies to BA.1 and BA.2 dropped significantly, with more profound changes observed in those without breakthrough Delta variant infection. Booster vaccination enhanced neutralising activities against ancestral strain and Delta variant, as compared to those induced by primary vaccination. These responses were maintained at high levels for at least 15 weeks. Our findings emphasise the importance of the first booster dose in producing cross-neutralising antibodies against Omicron variant. A second booster dose might be needed to maintain long-term protection against Omicron variant.

Transmission of SARS-CoV-2 Delta Variant Among Vaccinated Healthcare Workers, Vietnam (preprint)

Background: Data on breakthrough SARS-CoV-2 Delta variant infections are limited.Methods: We studied breakthrough infections among healthcare workers of a major infectious diseases hospital in Vietnam. We collected demographics, vaccination history and results of PCR diagnosis alongside clinical data. We measured SARS-CoV-2 (neutralizing) antibodies at diagnosis, and at week 1, 2 and 3 after diagnosis. We sequenced the viruses using ARTIC protocol.Findings: Between 11th–25th June 2021 (week 7–8 after dose 2), 69 healthcare workers were tested positive for SARS-CoV-2. 62 participated in the clinical study. 49 were (pre)symptomatic with one requiring oxygen supplementation. All recovered uneventfully. 23 complete-genome sequences were obtained. They all belonged to the Delta variant, and were phylogenetically distinct from the contemporary Delta variant sequences obtained from community transmission cases, suggestive of ongoing transmission between the workers. Viral loads of breakthrough Delta variant infection cases were 251 times higher than those of cases infected with old strains detected between March-April 2020. Time from diagnosis to PCR negative was 8–33 days (median: 21). Neutralizing antibody levels after vaccination and at diagnosis of the cases were lower than those in the matched uninfected controls. There was no correlation between vaccine-induced neutralizing antibody levels and viral loads or the development of symptoms.Interpretation: Breakthrough Delta variant infections are associated with high viral loads, prolonged PCR positivity, and low levels of vaccine-induced neutralizing antibodies, explaining the transmission between the vaccinated people. Physical distancing measures remain critical to reduce SARS-CoV-2 Delta variant transmission.Funding: Wellcome (106680/B/14/Z and 204904/Z/16/Z).Declaration of Interest: None to declare.Ethical Approval: The study was approved by the Institutional Review Board of HTD and the Oxford Tropical Research Ethics Committee, University of Oxford, UK.

Immunogenicity of Oxford-AstraZeneca COVID-19 vaccine in Vietnamese healthcare workers (preprint)

We studied the immunogenicity of Oxford-AstraZeneca vaccine in Vietnamese healthcare workers. We collected blood samples before each dose, at 14 days after each dose, and month 1 and 3 after dose 1 from each participant alongside demographics data. We measured neutralizing antibodies using a surrogate virus neutralization assay. The 554 study participants (136 males and 418 females) were aged between 22-71 years (median: 36 years). 104 and 94 out of 144 selected participants were successfully followed up at 14 days after dose 2 and 3 months after dose 1, respectively. Neutralizing antibodies increased after each dose, with the sero-conversion rate reaching 98.1% (102/104) at 14 days after dose 2. At month 3 after dose 1, neutralizing antibody levels decreased, while 94.7% (89/94) of the study participants remained seropositive. Oxford-AstraZeneca COVID-19 vaccine is immunogenic in Vietnamese healthcare workers. The requirement for a third dose warrants further research.

SARS-CoV-2 detection in nasopharyngeal throat swabs by metagenomics (preprint)

Metagenomics could detect SARS-CoV-2 in all eight nasopharyngeal/throat swabs with high/low viral loads, and rhinovirus in a co-infected patient. The sequenced viruses belonged to lineage B1. Because metagenomics could detect novel pathogen and co-infection, and generate sequence data for epidemiological investigation, it is an attractive approach for infectious-disease diagnosis.

How Digital Natives Learn and Thrive in the Digital Age: Evidence from an Emerging Economy

As a generation of ‘digital natives,’secondary students who were born from 2002 to 2010 have various approaches to acquiring digital knowledge. Digital literacy and resilience are crucial for them to navigate the digital world as much as the real world;however, these remain under-researched subjects, especially in developing countries. In Vietnam, the education system has put considerable effort into teaching students these skills to promote quality education as part of the United Nations-defined Sustainable Development Goal 4 (SDG4). This issue has proven especially salient amid the COVID−19 pandemic lockdowns, which had obliged most schools to switch to online forms of teaching. This study utilizes a dataset of 1061 Vietnamese students taken from the United Nations Educational, Scientific, and Cultural Organization (UNESCO)’s “Digital Kids Asia Pacific (DKAP)”project and employed Bayesian statistics to explore the relationship between the students’background and their digital abilities. Results show that economic status and parents’level of education are positively correlated with digital literacy. Students from urban schools have only a slightly higher level of digital literacy than their rural counterparts, suggesting that school location may not be a defining explanatory element in the variation of digital literacy and resilience among Vietnamese students. Students’digital literacy and, especially resilience, also have associations with their gender. Moreover, as students are digitally literate, they are more likely to be digitally resilient. Following SDG4, i.e., Quality Education, it is advisable for schools, and especially parents, to seriously invest in creating a safe, educational environment to enhance digital literacy among students.

The natural history and transmission potential of asymptomatic SARS-CoV-2 infection (preprint)

Background: Little is known about the natural history of asymptomatic SARS-CoV-2 infection or its contribution to infection transmission. Methods: We conducted a prospective study at a quarantine centre for COVID-19 in Ho Chi Minh City, Vietnam. We enrolled quarantined people with RT-PCR-confirmed SARS-CoV-2 infection, collecting clinical data, travel and contact history, and saliva at enrolment and daily nasopharyngeal throat swabs (NTS) for RT-PCR testing. We compared the natural history and transmission potential of asymptomatic and symptomatic individuals. Results: Between March 10th and April 4th, 2020, 14,000 quarantined people were tested for SARS-CoV-2; 49 were positive. Of these, 30 participated in the study: 13(43%) never had symptoms and 17(57%) were symptomatic. 17(57%) participants acquired their infection outside Vietnam. Compared with symptomatic individuals, asymptomatic people were less likely to have detectable SARS-CoV-2 in NTS samples collected at enrolment (8/13 (62%) vs. 17/17 (100%) P=0.02). SARS-CoV-2 RNA was detected in 20/27 (74%) available saliva; 7/11 (64%) in the asymptomatic and 13/16 (81%) in the symptomatic group (P=0.56). Analysis of the probability of RT-PCR positivity showed asymptomatic participants had faster viral clearance than symptomatic participants (P<0.001 for difference over first 19 days). This difference was most pronounced during the first week of follow-up. Two of the asymptomatic individuals appeared to transmit the infection to up to four contacts. Conclusions: Asymptomatic SARS-CoV-2 infection is common and can be detected by analysis of saliva or NTS. NTS viral loads fall faster in asymptomatic individuals, but they appear able to transmit the virus to others.